Channel confluences represent critical hydraulic control points governing sediment, nutrient, and pollutant transport in fluvial systems. While homogeneous confluence flows have been extensively studied, bottom-propagating gravity current intrusions driven by suspended sediment or thermal gradients remain inadequately resolved due to computational challenges in capturing multi-scale three-dimensional turbulent processes. This study employs high-resolution large-eddy simulations to systematically investigate intrusive gravity currents at a 90° channel confluence across velocity ratios (β) spanning 0 to 3.31 at Re = 58,200. The computational framework resolves buoyancy-momentum interactions using dynamic Smagorinsky subgrid-scale modeling with refined near-wall resolution (z⁺ ≤ 10). Simulations identify three distinct hydrodynamic regimes. For β ≤ 0.47 (buoyancy-dominated), currents exhibit symmetric radial spreading with coherent lobe-cleft patterns and vortex rings. The transitional regime (0.47 < β ≤ 1.42) demonstrates episodic deflection and intermittent vortex disruption. Beyond the critical threshold β = 1.42 (momentum-dominated), lateral spreading becomes suppressed to 0.2~0.25 channel widths, with rapid streamwise reorientation. Mass flux analyses quantitatively confirm the regime transition from perpendicular to longitudinal transport. The numerical results resolve interfacial coherent structure evolution from laminar sheet-like instabilities through Kelvin-Helmholtz roll-ups into tube-shaped and arch-like vortices, culminating in streamwise-oriented vorticity analogous to natural river confluences. Despite fundamental differences between density-driven and momentum-driven confluent flows, their interfacial vortical dynamics exhibit remarkable similarity, suggesting universal shear-driven mixing mechanisms. These high-fidelity simulations establish regime diagrams linking velocity ratios to transport patterns and provide validated insights for predicting complex hydraulic phenomena in natural and engineered systems.